Performance prediction of a liquid desiccant dehumidifier using artificial neural networks approach

Abstract

This work aims to develop two main ideas: first, the use of the artificial neural network (ANN) approach to predict the moisture removal rate (MRR) and the dehumidifier effectiveness (ɛ) of a counter-flow liquid desiccant dehumidifier using calcium chloride as an absorption solution. Second, the Garson method is used to identify the most important working parameters influencing the performance of the packed-bed dehumidifier component. A network model was developed in a MATLAB environment based on a multilayer perceptron that included an input, a hidden and an output layer. The network input parameters were the dry and wet bulb temperatures, the air and liquid flow rates, and the liquid desiccant temperature and concentration. The network output included two variables; the MRR and the ɛ. The performances of the ANN predictions were tested using experimental data not employed in the training process. The predicted values were found to be in good agreement with the experimental values, with mean relative errors of less than 4.90% for the MRR and 3.85% for ɛ. In addition, the air wet bulb and dry bulb temperatures were the parameters with the most influence on the MRR and ɛ, with a relative importance of 35% and 25%, respectively.